About hMetis comment section support

[wjrforcyber]

Why this

Hi!
I tried mt-kahypar on a file I generated by myself, I encountered with the

[hypergraph_io.cpp:readHypergraphFile:441]: Assertion `pos == handle.length` failed:

What

I use a file here:

i10s.txt

How

After I delete the comments section at the end, everything goes well, so maybe you are missing a skip line check on "%" before ASSERT(pos == handle.length)?

[larsgottesbueren]

FYI, the IDs in hMetis format are 1-based. The second line in your file has a zero.

[wjrforcyber]

FYI, the IDs in hMetis format are 1-based. The second line in your file has a zero.

Hi there!
Since what I dumped is from an AIG(AND-INV graph) representation, the index 0 indicates a const(which is not a real node, sometimes it does not have connections to others, but I treated it as a node in case it does have relationship with others), other index are real AND node with complemented edges.
Just curious, does your implementation have condition on this, or could you please show me on which section in the manual have restriction on 1-based node index? I checked on section 3.4 again maybe I miss it? I do see figure 5 start from 1 but don't see any explicit description on this.

[larsgottesbueren]

Not sure if it's documented. But trust the example in the manual! This fact is moderately well known.
Our implementation relies on this and will break. It will parse the zero correctly [0], but then subtract one [1] as intended, which leads to an underflow to numeric_limits<uint32_t>::max().

[0] /~https://github.com/kahypar/mt-kahypar/blob/master/mt-kahypar/io/hypergraph_io.cpp#L199
[1] /~https://github.com/kahypar/mt-kahypar/blob/master/mt-kahypar/io/hypergraph_io.cpp#L359